Classification of Chest XRay Diseases through image processing and analysis techniques

TL;DR

This work tackles automated interpretation of chest X-ray images for multi-class thoracic disease detection using DenseNet-121, EfficientNetV2, and Vision Transformer (ViT). It analyzes methods for handling uncertain labels (Label Powerset, Classifier Chains, Binary Mapping) and label smoothing within a four-step pipeline built on CheXpert, reduced to five clinically important classes. The study finds that a low learning rate $L$ of 0.0001 generally yields better convergence and F1 performance across architectures, with DenseNet-121 and EfficientNetV2 outperforming ViT in most settings. The results underscore the importance of hyperparameter tuning, data augmentation, and robust training regimes for reliable clinical support, and suggest future work involving larger datasets and expert validation. $L(y, \hat{y}) = - \sum_{i=1}^{C} y_i \log(\hat{y}_i)$ is used as the cross-entropy loss guiding model optimization; the work provides code at the linked repository for reproducibility.

Abstract

Multi-Classification Chest X-Ray Images are one of the most prevalent forms of radiological examination used for diagnosing thoracic diseases. In this study, we offer a concise overview of several methods employed for tackling this task, including DenseNet121. In addition, we deploy an open-source web-based application. In our study, we conduct tests to compare different methods and see how well they work. We also look closely at the weaknesses of the methods we propose and suggest ideas for making them better in the future. Our code is available at: https://github.com/AML4206-MINE20242/Proyecto_AML

Figures (2)

• Figure 1: Final approach for multi-classification task on the CheXPert Dataset
• Figure 2: Baseline architecture that consists of 4 dense blocks